Re: Defunct processes

vslabs_at_onwe.co.za (Billy Verreynne) wrote in message news:<1a75df45.0311240049.7969c471_at_posting.google.com>...
> Platform : HP-UX 11.00
> DB : Oracle 9.2.0
>
> On one platform I have PMON creating a fair size number (260+ last
> count) defunct processes. Do not see this behaviour on the other
> HP-UX/9i platforms.
>
> The PPID points to PMON thus I assume that these children were forked
> by PMON.
>
> Any ideas as to why these child processes are not being waited upon by
> PMON? Could this be caused by memory/swap issues? Some users run
> big/lots of processes exhausting swap and user memory... <sigh>
>
> I recall having problems some years ago on HP-UX with forking
> children. When a child terminated I did not always get a child died
> signal.. especially not when two child processes generated this signal
> at the same time (instead of 2 signals, the parent signal handler only
> received a single signal). Maybe something similar in this case
> causing PMON to remain in the dark about its child processes?

Sounds like you answered your own question. I'd guess running out of swap is preventing the children from saying "bye-bye." I take it you don't have a huge file-based secondary swap for these users? One of the reasons for a defunct process is that it is waiting on I/O, so if the I/O request cannot be met (ie, requesting and not getting swap space), the child will ignore everything.

Once they are defunct, I don't think PMON can do anything a kill -9 can't do. Which is nothing.

>
> Confirmation : these defunct processes do not consume any resources
> but a process struct in the process list table?

That's my understanding (man glossary /zombie), but you sure have a lot (of, um, persistent users? :-) . lsof http://ftp.cerias.purdue.edu/pub/tools/unix/sysutils/lsof/ might help, especially if you are being misdirected to look at pmon.

Oh, of course, check patches.

jg

--
@home.com is bogus.
http://www.signonsandiego.com/news/uniontrib/sat/metro/news_7m22ruling.html